Scientists challenge Google's quantum supremacy claims using 512 GPUs

Jimmy2x

Posts: 116   +8
Staff
Why it matters: In 2019, a paper leaked claiming physicists at Google used their quantum computer, Sycamore, to run a calculation that would overwhelm the world's most powerful supercomputer. Chinese scientists recently challenged that claim by successfully running the identical computation in a matter of hours using the computing power of today's GPUs. Their results prove a supercomputer using today's technology could likely beat Sycamore's previous record.

Google's quantum computing researchers originally ran the complex calculation in 200 seconds (just over three minutes), a feat they claimed would have taken the fastest supercomputer upwards of 10,000 years. Based on this result, the team claimed they had reached a significant milestone known as quantum supremacy. Quantum supremacy is the point at which a quantum device can solve problems that otherwise cannot be solved by classical technology in any reasonable amount of time.

Not everyone was a believer in Google's self-proclaimed supremacy in 2019. Another major player in the quantum computing space, IBM, challenged Google's claims from the start. Researchers there claimed the same task could be performed in a matter of days with the right amount of available resources, invalidating Google's claim of quantum supremacy.

Chinese scientists successfully proved IBM's point by attacking the original problem using advanced algorithms and compute power from today's GPUs to complete the calculation. A report in Science notes that the effort used 512 GPUs — a number that is far from unfathomable when considering how many units cryptocurrency mining operations sometimes use.

Also see: What is Quantum Computing?

The GPU compute power combined with advanced algorithms completed the same calculation within several hours. The results, which were unthinkable according to the leaked 2019 research findings, provide evidence to back claims that a large enough supercomputer could, in fact, rival Sycamore's earlier achievement.

Conventional computing relies on bits, the most basic information units in computing. These bits can exist as only one of two values, either a 0 or 1.

Quantum computing relies on quantum bits, or qubits, comprised of a superposition of 0 and 1. Like a bit, a qubit can equal 0 or 1. However, it has the added property of equaling 0 and 1 simultaneously, resulting in vastly increased computing potential.

The achievement does not invalidate Google's previous quantum achievements, nor does it mean that standard processing hardware can "catch up" to quantum's capabilities. Google Quantum AI's principal scientist, Sergio Boixo, said the original 2019 paper acknowledged the likely future improvement to classical algorithms but doesn't believe today's classical computing approach can keep pace with quantum technologies.

Boixo's statement is accurate given the rate of quantum growth since 2019. Google's original Sycamore was a 53-qubits processor. In 2021 IBM unveiled their 127-qubit Eagle, and their quantum roadmap looks to break the 1,000-qubit barrier sometime in 2023.

Permalink to story.

 

Tantor

Posts: 349   +629
"However, it has the added property of equaling 0 and 1 simultaneously, resulting in vastly increased computing potential."

LOL. This is like saying that a chicken can be a hen and rooster simultaneously, resulting in vastly increased egg laying potential.

Note that quantum is no longer binary. It adds two more states, simultaneity and non-simultaneity. It's also analog, not digital, because these states phase in and out. The rate of phasing adds even more complexity. Under Quantum computing laws, it's possible to show that a cluster of 4 qbits = 5, or 2+2=5. This is due to qbit resonance.

Quantum computing is to math what transgender is to sexuality.

Quantum computing is limited by the difficulty of discriminating between individual qbit states on a large scale. It's like tracking individual water molecules in a a flowing stream. As the Chinese showed, the strength of conventional computing is that it can create massive arrays of much slower computing units.
 

kiwigraeme

Posts: 1,202   +872
Google - being dishonest - what a surprise.

Did you actually read the article so end result google 200 seconds - China 512 GPUs a couple of hours
Googles claim still technically not overturned using established algorithm .
China used new advanced algorithm as IBM claimed was possible ( note IBM failed for 2 years to prove it )

So not the big gotcha

There was story the other day of a new encryption broken quite quickly despite the makers claims -as they thought it was independent of a mathematic technique that did not help earlier encryption methods
This is the nature of science - many problems that take 1000s of years to compute how been cracked using new tools .
Fundamental research by Google is done to a pretty high std .
When it comes to quantum computers it's that Canadian D-wave that seems to make outlandish claims
META , Google, Microsoft , IBM ( do they have most patents from memory ) all do a high std of basic research - so I will bring in my usual rant Apple does nothing to promote open science - and it's completely selfish

There is a lot of bragging rights amongst scientists - so great IBM pulled them up - means those scientists from both companies will work harder

QC like Fusion is worth a lot of effort - whether we will get something effective soon is a great unknown
 

OortCloud

Posts: 796   +777
Did you actually read the article so end result google 200 seconds - China 512 GPUs a couple of hours
Yes I did. Google claimed it would take the worlds fastest supercomputer 10,000 years to solve this. Given china did it with 512 GPUs in a couple of hours I think that it is probably unlikely it would take the worlds fastest supercomputer 10,000 years don't you? Sure 512 GPUs is a lot but these supercomputers are no slouches either. The current head of the table is believed to be Perimutter at the Department of Energy. As just part of its arsenal It features 6,159 Nvidia A100 GPUs. I'm guessing it would brute force through this problem in minutes making Googles 10,000 years claim even more ludicrous.
 

kiwigraeme

Posts: 1,202   +872
Yes I did. Google claimed it would take the worlds fastest supercomputer 10,000 years to solve this. Given china did it with 512 GPUs in a couple of hours I think that it is probably unlikely it would take the worlds fastest supercomputer 10,000 years don't you? Sure 512 GPUs is a lot but these supercomputers are no slouches either. The current head of the table is believed to be Perimutter at the Department of Energy. As just part of its arsenal It features 6,159 Nvidia A100 GPUs. I'm guessing it would brute force through this problem in minutes making Googles 10,000 years claim even more ludicrous.
Yeah I get that - but google were assuming the standard algorithm - so their statement for that might be true - no one has claimed otherwise .
IBM suggested a more advanced algorithm would knock it off much sooner .
Which these chinese researchers showed

These articles probably miss a whole lot
what likely happened in my view
Google scientists - hey you may be right - but prove it IBM
a couple of months later Google internally - we will need something else to show quantum supremacy - I think IBM is right - about new techniques .
Mathematics won't be their forte - though they will have mathematicians and they probably speak with top Mathematicians .

So this example may have entirely have been suggested years before as a good test - and no one disagreed in principle.

ie I don't think we know the full story - thinks aren't static like this article implied

I agreed that Google has BS about a lot of things - but I haven't seen this really on the research side - just on the product , data mining , advertising , browser algorithm etc

I don't think they even may huge claims about their driverless cars -happy to be proved wrong
 

kiwigraeme

Posts: 1,202   +872
Yes I did. Google claimed it would take the worlds fastest supercomputer 10,000 years to solve this. Given china did it with 512 GPUs in a couple of hours I think that it is probably unlikely it would take the worlds fastest supercomputer 10,000 years don't you? Sure 512 GPUs is a lot but these supercomputers are no slouches either. The current head of the table is believed to be Perimutter at the Department of Energy. As just part of its arsenal It features 6,159 Nvidia A100 GPUs. I'm guessing it would brute force through this problem in minutes making Googles 10,000 years claim even more ludicrous.

Given your handle - Supercomputers kind of like travelling to the nearest star.
When you finish building one , is obsolete to a new design
So you send a spaceship to the next star by std rocket and slingshot
Then you send a nuclear powered spaceship
Then a solar sail spaceship powered by lasers
Then a ramjet
Then a fusion drive
Then antimatter drive

and they all arrive in the order of last one sent is first
 

OortCloud

Posts: 796   +777
I agreed that Google has BS about a lot of things - but I haven't seen this really on the research side - just on the product , data mining , advertising , browser algorithm etc
Yeah I really dislike Google - the whole model of basically giving away free operating systems, search engines and mobile OS's is inherently massively dishonest. Nothing is for free and with Google you are being scraped in ways you never thought possible. The adwords API's really highlight this. Ten minutes after looking up 'new lawnmowers' you suddenly start getting adverts for them on websites. If that were the only way they scrape your data I might be able to live with it, but it's much more pernicious than that. The profiles they build for adwords are a real invasion of privacy and avoiding it is always difficult/impossible. Click here to see the tip of the iceberg on their data they store on you...


But I agree Quantum computing is an incredible technology and given the right problems can be incredibly powerful. I just wish it was MIT and CalTech say leading the charge, not these huge corporations who will just use it to build better models on us and sell us more stuff we don't need.
 
Last edited:

human7

Posts: 69   +48
TL;DR: Google exaggerated, which is what many people thought at the time (including myself). That doesn't invalidate the achievement though - 3 minutes vs several hours is a big difference, and the result from 2019 still proves that quantum computers have enormous potential (not that anyone was doubting that). Even if classical computers can do the computation in 3 minutes, Google's quantum computer at the time only had 53 qubits. While I don't know what was needed to keep that computer operational, I imagine it was far more power efficient than the classical systems.

In any case, the computational power of a quantum computer increases exponentially as the number of qubits increase (although, so do the chances for errors, hence the strong need for quantum ECC). We'll still need absolutely incredible quantum computers to break RSA in a reasonable time, but maybe lots of other problems can be solved quickly in the meantime.
 

Hodor

Posts: 113   +67
Now, what happens when they start using quantum computers to mine bitcoins and other crypto? They'll be able to pile up huge amount of "money" with minimal energy consumption, hence abusing the whole crypto system.
 

kiwigraeme

Posts: 1,202   +872
Yeah I really dislike Google - the whole model of basically giving away free operating systems, search engines and mobile OS's is inherently massively dishonest. Nothing is for free and with Google you are being scraped in ways you never thought possible. The adwords API's really highlight this. Ten minutes after looking up 'new lawnmowers' you suddenly start getting adverts for them on websites. If that were the only way they scrape your data I might be able to live with it, but it's much more pernicious than that. The profiles they build for adwords are a real invasion of privacy and avoiding it is always difficult/impossible. Click here to see the tip of the iceberg on their data they store on you...


But I agree Quantum computing is an incredible technology and given the right problems can be incredibly powerful. I just wish it was MIT and CalTech say leading the charge, not these huge corporations who will just use it to build better models on us and sell us more stuff we don't need.
I follow that link - got my sex and age in right ball park - said I was married - I suppose a 18 year defacto is that . Said my income was upper middle - kind of - I retired 10 years early and living on so-so investment income - I have no real expenses like rent/mortgage - some areas frugal , some spend .
Said company 250-10000 - I haven't worked corporate for 30 years - ran my own small business.
Then listed what seem a huge range of interests over 120. It hit the big ones - some funny off kilter ones - eg hatchbacks .
I have a huge curiosity about the world - plus if I help someone buy something of note - like researching what to buy.
No mention of my sexuality- Believe META does that - but I don't do facebook , IG, SC, Twitter etc - just some reddit and some forums - and discord only if have to get some info ( so once every 3 months - as that's how that person posts ) .
Don't really mind google scraping me - even at Uni - I did Business Admin - I hated all the networking crap - made me puke - smarmy insincere un-interesting venal people .
Reminds me speaking to an ex-addict - he said one of the worse things about being one - was sucking up to dealers who were people you disliked .
As for YT if I look at a conspiracy nuts video - then I have to do - "don't show me this channel again " and the others that follow - same if clickbait ones come in my feed , or any video with a computer generated voice - can't stand those - automated BS content generators
 

Bobbydpue

Posts: 371   +246
Yes I did. Google claimed it would take the worlds fastest supercomputer 10,000 years to solve this. Given china did it with 512 GPUs in a couple of hours I think that it is probably unlikely it would take the worlds fastest supercomputer 10,000 years don't you? Sure 512 GPUs is a lot but these supercomputers are no slouches either.
Google made its claim in 2019 and at the time their claim was only challenged with claims and not practical examples. 200 seconds versus "several hours" is still impressive especially after 3 years of technical advancement.

I want to point out Chinese group that did this experiment wasn't named and neither was the exact amount of time it took to complete the "computation".
 

Bobbydpue

Posts: 371   +246
I hated all the networking crap - made me puke - smarmy insincere un-interesting venal people .
Reminds me speaking to an ex-addict - he said one of the worse things about being one - was sucking up to dealers who were people you disliked .
I don't work with "smarmy insincere people", but networking isn't sucking up to people you don't like. Meeting people in person and getting to know them changes relationships from some-guy-in-an-mail into and actual person. I've watched my working relationship go from sending an email and making a phone call and getting nowhere to getting on the phone having a quick conversation about their spouse or kids and then getting real progress on previously stalled projects.

People don't put as much effort into helping nameless faces as they do for people they've gotten to know and shared a meal or two with. There is a reason why companies take their clients out for a meal, because it is effective.
 

wiyosaya

Posts: 7,958   +7,001
Did you actually read the article so end result google 200 seconds - China 512 GPUs a couple of hours
Googles claim still technically not overturned using established algorithm .
I disagree.

Since Google's claim is essentially that the problem was unsolvable using conventional hardware, the claim was overturned. Even though it took conventional hardware a couple of hours to solve the problem, nevertheless, it was solved with conventional hardware - much to Google's disdain, I imagine.
 

Kirby1

Posts: 174   +283
Google - being dishonest - what a surprise.
I wouldn't call it dishonest. If anything, the term "Quantum Supremacy" is probably not well defined. It took three years and a cluster of 512 GPUs to be able to do in several hours what google did in just a few minutes.
 

Kirby1

Posts: 174   +283
Now, what happens when they start using quantum computers to mine bitcoins and other crypto? They'll be able to pile up huge amount of "money" with minimal energy consumption, hence abusing the whole crypto system.
It wouldn't be profitable. The reason mining works is because the hardware to do it is affordable and mass produced.
 

Gregtombstone

Posts: 29   +18
While the Chinese matched Google they probably used 125 kw of power. I'm guessing 500 gpu for 4 hours at 250 watts per gpu. Cant see a Google machine using that sort of power for a few seconds of computing unless you factor in the super cooling overhead. Effectively Quantum computing can save a huge amount of energy. Here in Ireland we are looking at power cuts this winter as no one calculated the load on our Grid of allowing 50 plus large server farms too be built here. There are more on the way and its reckoned they will use 30% of Ireland energy out put per year by 2030, 2 giga watts plus. Aside from Quantum computing Spintronics , where single electrons are given an up / down "spin" will save huge amounts of energy. Currently (no pun intended) it takes a train of 200/300 electrons flowing down a conductor to register . With a spun electron only 1 is needed to change a switch / gate a huge energy saving and reduced cooling requirements.
 

kiwigraeme

Posts: 1,202   +872
I don't work with "smarmy insincere people", but networking isn't sucking up to people you don't like. Meeting people in person and getting to know them changes relationships from some-guy-in-an-mail into and actual person. I've watched my working relationship go from sending an email and making a phone call and getting nowhere to getting on the phone having a quick conversation about their spouse or kids and then getting real progress on previously stalled projects.

People don't put as much effort into helping nameless faces as they do for people they've gotten to know and shared a meal or two with. There is a reason why companies take their clients out for a meal, because it is effective.

Yeah I understand networking is fine - we all do it to some extent . It was just the super insincere people - who don't really care about you - just who you are and what you can do for them - made worse if they treat not useful people poorly ( thankfully even a smaller subset )
I started backpacking the world in 87 having landed in Argentina - I had visions of meeting all these other tourists , collecting addresses - yay great he lives in NY , she in Berlin etc . A couple of weeks later I realised I only wanted addresses of people I liked and wanted to see again.

So you are right networking done right is a good thing

I came from a rough area of town - insincere people were a rarity . Get to Uni- yeah I be there of Saturday to help out ( no intention of coming - too scared to say F off I have better things to do )
 

bviktor

Posts: 957   +1,401
Now, what happens when they start using quantum computers to mine bitcoins and other crypto? They'll be able to pile up huge amount of "money" with minimal energy consumption, hence abusing the whole crypto system.
Nothing happens. PoW is stupidly complex _by choice_, and not because verifying a transaction requires that much computing power.

That's the whole reason why PoW needs to die a quick death - it's a complete waste of resources.